Professional Documents
Culture Documents
By Tom Irvine
Email: tomirvine@aol.com
February 8, 2012
_______________________________________________________________________________
This paper applies Smallwood’s methodology for base excitation in Reference 1 to the case of an
applied force. It also extends the method to multi-degree-of-freedom systems.
Ahlin & Brandt purported in Reference 2 to use the method given in the present paper, but they
omitted a derivation and the resulting filtering coefficients.
This paper fills the gap and gives accompanying Matlab scripts. The goals are to publicize the
method and to place the filter coefficients equations in the public domain.
Introduction
The purpose of this paper is to derive an efficient and accurate method for the modal transient
analysis of a dynamic system subjected to an external force or forces. The method will be a digital
recursive filtering relationship using the ramp invariant technique which models the slope between
adjacent points of the input force.
Assumptions
A modal analysis of the system has been previously performed. The system has been reduced to
uncoupled mass, damping and stiffness matrices per the method given in Reference 3, as well as in
common structural dynamics textbooks. The physical responses can then be recovered from the
modal responses after the transient analysis.
The initial conditions are zero. Note that the response to initial conditions can be solved exactly
using Laplace transforms, as needed.
Forcing Function
The forcing function may be either from measured data1 or from a synthesis.
1
Measured data should be collected using the proper sample rate and analog anti-aliasing filter.
1
The time step should be chosen so that there are at least ten points per cycle for the highest natural
frequency of interest. In other words, the sample rate should be at least ten times the highest
natural frequency of interest. This is an industry standard rule-of-thumb for time domain
calculations per Reference 4.
Smallwood wrote in his seminal paper that the ramp invariant method allows for analysis
frequencies approaching the Nyquist frequency, which in one-half the sample rate. But the x10
rule is still recommend in the present paper.
Note that the time step and corresponding sample rate are fixed for measured data as the data is
being collected. The time step can be shortened via interpolation or a cubic spline fit as a post-
processing step, but caution must be exercise in either case.
The time step can be set as small as needed for the case where a function is to be synthesized.
Alternatively, the number of modes included in analysis can be truncated if needed to meet the
rule-of-thumb. The highest frequency of interest is thus lowered as a trade-off.
Candidate Methods
Convolution Integral
The Convolution method would yield an exact answer for the response if the system and the
forcing function could be analyzed in analog rather than digital form. This is impractical,
however.
The nested series representation of the Convolution integral is not commonly used because it is
numerically inefficient.2
The Convolution integral represented as digital cursive filtering relationship. The derivation is
performed using a Z-transform.
2
Convolution can also be performed in the frequency domain my by multiplying the Fourier
transform of the applied force by the frequency response function of the system. The inverse
Fourier transform of the product gives the response time history. A Fast Fourier transform (FFT)
and its corresponding inverse can be used to expedite the calculation. Note that there are some
potential error sources with this method such as leakage.
2
There are two Z-transform approaches: the impulse invariant and ramp invariant simulations. The
ramp invariant simulation is preferred because it adds a filtering term and has better accuracy at
frequencies approach the Nyquist frequency per Reference 1.
Note that either digital recursive filtering relationship requires a constant time step.
Newmark-beta Method
The Newmark-beta method is derived from the continuous time equation familiar to high school
physics students.
1
u u t u t 2 (0)
2
It is typically implemented with 1 / 4 , thus assuming constant average acceleration over the
time step.
Runge-Kutta Method
The Runge-Kutta method extends the Taylor series method by estimating higher order derivatives
at points within the time step.
There are many types of Runge-Kutta algorithms. A common type which may be used for an
arbitrary forcing function is the Runge-Kutta fourth order method. This method is given in
Reference 7.
But the Runge-Kutta fourth order method may give unstable results3 for stiff systems with higher
natural frequencies.
The probability that instability will occur is difficult to predict. It depends on both the natural
frequency and the time step. The problem can be mitigated by using a smaller time step, but this
requires a longer analysis time.
If an instability occurs, then the analysis be repeated using a fewer number of modes for the case
of a multi-degree-of-freedom system.
3
“To infinity and beyond,” recalling Buzz Lightyear.
3
SDOF Equation of Motion
x
k c
Figure 1.
m is the mass
c is the viscous damping coefficient
k is the stiffness
x is the absolute displacement of the mass
f(t) is the applied force
x
m
kx c x
Figure 2.
4
Summation of forces in the vertical direction
F
mx (1)
cx kx f ( t )
mx (2)
cx kx f ( t )
mx (3)
Divide through by m,
c k 1
x x x f ( t ) (4)
m m m
By convention,
(c / m) 2n (5)
(k / m) n 2 (6)
where
By substitution,
1
x 2 n x n2 x f (t) (7)
m
Equation (7) does not have a closed-form solution for the general case in which f(t) is an arbitrary
function. A convolution integral approach must be used to solve the equation.
The solution method proceeds by finding a solution to the homogeneous form of equation (7). In
other words, the solution is found for f(t)=0.
x 2 n x n2 x 0
(8)
Note that equation (8) is essentially the same as equation (A-1) in Reference 8, except that
equation (8) is expressed in terms of absolute displacement.
5
Displacement
d n 1 - 2 (9)
ĥ d ( t ) =
1
exp n t sin d t (11)
md
1 1
H d (s) = (12)
m s 2 2 s 2
n n
1 z 12
Z L1
1
Ĥ d (z) =
m Tz s 2 s 2 2n s n 2
(13)
6
Evaluate the inverse Laplace transform per References 10 and 11.
2
1 n 2 sin t
L1
1 2
2n n t 2n cos d t 1 2 d
s 2 s 2 2n s n 2 n4
d
(14)
Ĥ d (z) =
z 12
2
n
Z 2n n t exp n t 2n cosd t 1 2 sin d t
1 2 2
4 Tz
mn d
(15)
Let
1 22
2
n
(16)
d
2n (17)
7
The Z-transform is evaluated using the method in Reference 12.
- z 12 z
Ĥ d (z)
m n Tz z 1
4
2
n z 12 T z
4 Tz z 12
m n
(18)
Ĥ d (z)
1
4
z 1 n2
m n T
z 12
1 z exp n T cosd T
4 T z 2 2zexp T cos T exp 2 T
m n n d n
8
Ĥ d (z)
1
4
z 1 n2
mn T
(20)
Ĥ d (z)
1 z 2 T
4 n
m n T
1
4
z 12 z 2 z2zexp
exp n T sin d T cosd T
T cos T exp 2 T
m n T n d n
(21)
Let
2
n T (25)
By substitution,
Ĥ d (z)
1
4
z 1
4
z 12 z 2zz (26)
mn T mn T
Ĥ d (z)
1
4
2
z z
z z 2 z 1
4
z 2 2z 1 z 2zz (27)
mn T mn T
9
2
Ĥ d (z)
1
z z 2 z
m n T
4 z z
2
1
z 2 z
m n T
4 z z
1
4
z 2 z 2z z
m n T
z
1
2z
m n T
4 z 2 z
1 z
4 z 2 z
m n T
(28)
1 z 3 z 2 z 1 z 2 z
Ĥ d (z)
mn T mn4 T z 2 z
4 z 2 z
1 z 3 z 2 1 2z 2 2z 1 z
mn T z z mn T z z mn T z z
4 2 4 2 4 2
(29)
2z 2 2 z
Ĥ d (z) (31)
4
mn T z 2 z
10
Solve for the filter coefficients using the method in Reference 1.
b 0 z 2 b1 z b 2
2z 2 2 z / mn4T
z 2 a1z a 2 z 2 z
(32)
Solve for a1.
Note that the a1 and a2 coefficients are common for displacement, velocity and acceleration.
Solve for b0.
b 0 2 / mn T
4
(35)
exp n T sin d T cosd T n T 2 exp n T cosd T 2
2
b0
4
mn T
(37)
11
b0
2
exp n T
n
1 2 sin d T 2n cosd T
2
d
4
m n T
2
exp n T n 1 2 2 sin d T n T 2n exp n T cosd T 1
2
b0 d
4
m n T
(40)
exp n T
n
2
1 2 sin d T n T 2exp n T cosd T 1
b0 d
3
m n T
(41)
2exp n T cosd T 1 exp n T
n
2 1 sin d T n T
2
b0 d
3
m n T
(42)
12
Solve for b1.
2
b1 (43)
4
mn T
2 1
b1 (44)
4
mn T
b1
2 n T exp n T cosd T 2 exp n T sin d T cosd T
2
4
mn T
2n T exp n T cosd T 2 exp n T sin d T 1 exp 2n T
2
b1
4
mn T
(46)
2n T exp n T cosd T 2 exp n T sin d T 1 exp 2n T
2
b1
4
mn T
(47)
13
b1
2
2 n T exp n T cosd T 2 n 1 2 2 exp n T sin d T
2
d
4
m n T
b1
2 n T exp n T cosd T 21 exp 2n T 2 n 2 2 1 exp n T sin d T
d
3
m n T
(49)
Solve for b2.
b2 (50)
4
mn T
14
b2
2
n n
n n
2 2 T exp 2 T exp T n 1 2 2 sin T 2 cos T
d n d
d
4
m n T
(52)
n
n
2 T exp 2 T exp T n 2 2 1 sin T 2 cos T
n d d
d
b2
3
m n T
(53)
xi a1 x i 1 a 2x i2
b0 f i b1 f i 1 b 2 f i 2
(54)
15
The digital recursive relationship for the displacement is thus
xi
exp 2n t x i 2
1
3
2exp n T cosd T 1 exp n T
n
2
2 1 sin d T n T f i
m n T d
1
3
2 n T exp n T cosd T 21 exp 2n T 2
n
2 1 exp n T sin d T f i 1
2
m n T d
1
3
n
2 n T exp 2n T exp n T
2
2 1 sin d T 2 cosd T f i 2
m n T d
(55)
16
Velocity
ĥ v ( t ) =
1
exp(- n t ) n
sin d t cos d t (56)
m d
1 s
H v (s) = (57)
m s 2 2 s 2
n n
1 z 12 1 s
Z L
Ĥ v (z) = (58)
m Tz s 2 s 2 2n s n 2
1 z 12
Z L1
1
Ĥ v (z) = (59)
m Tz s s 2 2n s n 2
L1 exp n t cosd t n sin d t
1 1 1
(60)
s s 2 2n s n 2 n2 n2 d
17
z 12
n
Z 1 exp n t cosd t sin d t
1
Ĥ v (z) = (62)
2 Tz
m n d
n
z z exp( T ) cos( T ) z exp t sin d
t
1 z 12 z
n d n
d
Ĥ v (z) =
mn Tz z 1
2
z 2 z exp( n T) cos(d T) exp 2n T
2
(63)
n
z z exp( T ) cos( T ) z exp t sin d
t
1 z 12 z
n d n
d
Ĥ v (z) =
mn T z z 1
2 z 2 2 z exp( n T) cos(d T) exp 2n T
(64)
z exp( n T) cos(d T) n exp n t sin d t
1 d
Ĥ v (z) =
2
z 1 z 12
mn T z 2 z exp( n T) cos(d T) exp 2n T
2
(65)
18
z exp( n T) cos(d T) n sin d t
1 d
Ĥ v (z) = z 1 z 12
2
m n T z 2 z exp( n T) cos(d T) exp 2n T
2
(66)
Let
exp( n T) cos(d T) n sin d t (67)
d
z
z 1 z 12
1
Ĥ v (z) = (70)
mn T z 2 z
2
Ĥ v (z) =
1
z 1 z 2 2z 1 z
(71)
mn T z 2 z
2
Ĥ v (z) =
1
z 1
z z 2 2z 1 z 2 2z 1
(72)
mn T
2 z2 z
Ĥ v (z) =
1
z 1
z 3 2z 2 z z 2 2z (72)
mn T
2 z2 z
z 3 2z 2 z z 2 2z
z 1
1
Ĥ v (z) = (74)
mn T
2 z2 z
19
z 3 2 z 2 1 2 z
z 1
1
Ĥ v (z) = (75)
mn T
2 z2 z
z 3 2 z 2 1 2 z
z 1
1
Ĥ v (z) = (76)
mn T
2 z2 z
z 2 z z 3 2 z 2 1 2 z
z 1
1
Ĥ v (z) = (77)
mn T
2 z2 z z2 z
1 z 3 z 2 z z 2 z z 3 2 z 2 1 2 z
Ĥ v (z) = (78)
2 2
z z 2
z z
mn T
z 3 1 z 2 z z 3 2 z 2 1 2 z
1
Ĥ v (z) = (79)
mn T
2 z2 z z2 z
1 2 1z 2 1 2 z
Ĥ v (z) = (80)
mn T
2 z2 z
1 1z 2 2 1z
Ĥ v (z) = (81)
2 2 z
mn T z
20
1 1z 2 2 1z
Ĥ v (z) = (82)
2 2
z z
mn T
c 0 z 2 c1 z c 2 1 1z 2 2 1z
= (83)
z 2 a1z a 2 2
mn T z2 z
1
c0 (86)
2
m n T
exp( n T) cos(d T) n sin d t 2 exp n T cosd T 1
d
c0 (87)
2
mn T
exp( n T) cos(d T) n sin d t 1
d
c0 (88)
2
mn T
21
Solve for c1.
2 1
c1 (89)
2
m n T
exp 2n T 2 exp n T cosd T 2 exp( n T) cos(d T) n sin d t 1
d
c1
2
mn T
(90)
exp 2n T 2 exp( n T) n sin d t 1
d
c1 (91)
2
mn T
c2 (92)
2
mn T
exp( n T) cos(d T) n sin d t exp 2n T
d
c2 (93)
2
mn T
x i a1 x i 1 a 2 x i 2
c0 f i c1 f i 1 c 2 f i 2
(94)
22
The digital recursive relationship for the velocity is thus
x i
n
sin d t 1 f i
1
exp( n T) cos(d T)
2
m n T d
n sin t 1 f
1
2
exp 2n T 2 exp( n T ) d i 1
m n T d
n
1
2
exp( n T ) cos( d T ) sin d t exp 2
n f i2
T
m n T d
(95)
Acceleration
yt
1
m
( t ) exp n t 2n cosd t
n 2
d
2 2 1 sin d t
(96)
1 s2
H a (s) = (97)
m s 2 2 s n 2
n
The Z-transform is
1 z 12 1 s2
Z L 2 2
Ĥ a (z) =
m Tz
s s 2 s n 2
n
(98)
23
1 z 12 1 s2
Z L 2 2
Ĥ a (z) =
m Tz
s s 2 s n 2
n
(99)
1 z 12 1 1
Ĥ a (z) = Z L (100)
m Tz s 2 2 s n 2
n
1 1
L1 L1 (101)
s n d
s 2 s n
2 2 2 2
n
L 1
1
s 2 s n
2 2
1
exp n t sin d t (102)
n d
1 z 12 1
exp n t sin d t
Ĥ a (z) = Z (103)
m Tz
d
1 z 12 1
Ĥ a (z) =
Z exp n t sin d t
(104)
m Tz
d
24
1 z 12 exp n t sin d t
Ĥ a (z) = (106)
md T z 2 z exp( n T) cos(d T) exp 2n T
2
(109)
25
Solve for d1.
d1 2d 0 (113)
d2 d0 (114)
If two are known, the third can be calculated via equation (7).
26
SDOF Example
An SDOF system has a natural frequency of 10 Hz with an amplification of Q=10. Its mass is 1
lbm. It is subjected to 1 lbf sinusoidal excitation at is natural frequency as shown in Figure 3.
The analysis is performed using the ramp invariant digital recursive filters derived in this paper as
implemented in Matlab script: arbit_force.m.
This is a simple problem which does not demonstrate the full power of the ramp invariant filtering
method for calculating the response to a force which varies arbitrarily with time.
But the simple example is useful for checking the accuracy of the method. The peak results agree
with the expected values from the corresponding frequency response functions.
The descriptive statistics for the response from the Matlab script are:
Displacement Response
Velocity Response
Acceleration Response
maximum = 9.98 G
minimum = -9.97 G
overall = 6.17 G RMS
27
APPLIED FORCE
1.0
0.5
FORCE (lbf)
-0.5
-1.0
0 0.5 1.0 1.5 2.0
TIME (SEC)
Figure 3.
28
SDOF DISPLACEMENT RESPONSE fn=10 Hz Q=10
1.2
Laplace
1.0 Ramp Invariant
0.8
0.6
0.4
0.2
DISP (inch)
-0.2
-0.4
-0.6
-0.8
-1.0
-1.2
0 0.5 1.0 1.5 2.0
TIME (SEC)
Figure 4.
The resulting displacement, velocity and acceleration responses for the ramp invariant method are
shown in Figures 4, 5 and 6, respectively. The exact results from the Laplace transform
calculation are superimposed for comparison, as calculated from the formulas Reference 13 as
implemented via Matlab script: sdof_sine_force.m.
There is excellent agreement between the two displacement curves as shown in Figure 4.
29
SDOF VELOCITY RESPONSE fn=10 Hz Q=10
100
Laplace
Ramp Invariant
80
60
40
VELOCITY (inch/sec)
20
-20
-40
-60
-80
-100
0 0.5 1.0 1.5 2.0
TIME (SEC)
Figure 5.
There is very good agreement between the two velocity curves as shown in Figure 4.
The Ramp Invariant curve is slightly ahead of the Laplace curve in terms of phase. This
difference will be investigated in the next revision of this paper.
30
SDOF ACCELERATION RESPONSE fn=10 Hz Q=10
14
Laplace
12 Ramp Invariant
10
4
ACCEL (G)
-2
-4
-6
-8
-10
-12
-14
0 0.5 1.0 1.5 2.0
TIME (SEC)
Figure 6.
There is excellent agreement between the two acceleration curves as shown in Figure 6.
31
MDOF Application
The ramp invariant digital recursive filtering relationship can be readily used as the numerical
engine in an MDOF modal transient analysis for each of the respective response parameters.
An example is shown in Appendix A.
References
2. A. Brandt & K. Ahlin, A Digital Filter Method for Forced Response Computation, Society
for Experimental Mechanics ( SEM ) Proceedings, IMAC-XXI.
3. T. Irvine, The Generalized Coordinate Method for Discrete Systems, Revision F,
Vibrationdata, 2012.
4. Himelblau, Piersol, et al., IES Recommended Practice 012.1: Handbook for Dynamic Data
Acquisition and Analysis, Institute of Environmental Sciences and Technology, Mount
Prospect, Illinois.
5. Rao V. Dukkipati, Vehicle Dynamics, CRC Press, Narosa Publishing House, New York.
2000.
11. T. Irvine, Partial Fractions in Shock and Vibration Analysis, Revision I, Vibrationdata,
2012.
32
APPENDIX A
MDOF Example
The example from Reference 2 is repeated here.
k3 f2(t)
x2
m2
k2 f1(t)
x1
m1
k1
Figure A-1.
33
The parameters are
B1 100 lbf
B2 200 lbf
55 Hz
100 Hz
m 0 3 0
M 1 (A-1)
0 m 2 0 2
k k 2 k 2 700,000 300,000
K 1
k 2 k 3 300,000 400,000
(A-2)
k2
Each of the two forcing functions is synthesized using Matlab script: generate.m, as a pre-
processing step.
34
>> mdof_modal_arbit_force_ri
Mass Matrix
Enter the matrix name: mass_case5
Stiffness Matrix
Enter the matrix name: stiff_case5
Natural Frequencies
No. f(Hz)
1. 48.552
2. 92.839
ModeShapes =
0.3797 -0.4349
0.5326 0.4651
Enter duration(sec)
35
0.3
Each force file must have two columns: time(sec) & force(lbf)
Calculating response...
36
DISPLACEMENT
0.003
dof 2
dof 1
0.002
0.001
DISP (INCH)
-0.001
-0.002
-0.003
0 0.05 0.10 0.15 0.20 0.25 0.30
TIME (SEC)
Figure A-2.
The displacement, velocity and acceleration responses are shown in Figures A-2, A-3 and A-4,
respectively.
The results appear to be the same as the Laplace transform results in Reference 3. A formal
comparison will be given in the next revision of this paper.
37
VELOCITY
1.5
dof 2
dof 1
1.0
0.5
VEL (INCH/SEC)
-0.5
-1.0
-1.5
0 0.05 0.10 0.15 0.20 0.25 0.30
TIME (SEC)
Figure A-3.
38
ACCELERATION
1.5
dof 2
dof 1
1.0
0.5
ACCEL (G)
-0.5
-1.0
-1.5
0 0.05 0.10 0.15 0.20 0.25 0.30
TIME (SEC)
Figure A-4.
39